Hsnet-based signal-to-noise ratio robust isar imaging method and device

Abstract

The application provides a signal-to-noise ratio robust ISAR imaging method and device based on an HSNet. The method comprises the following steps: obtaining ISAR echo data to be imaged; performing motion compensation and range-Doppler imaging processing on the ISAR echo data to be imaged to obtain a range-Doppler image of original echo; inputting the range-Doppler image of the original echo into a pre-trained HSNet network to output parameter information of a reconstructed ISAR image; obtaining an ISAR imaging result based on the parameter information of the reconstructed ISAR image; and setting a preset loss function of the pre-trained HSNet network as a negative variational lower bound. In the application, the parameter information of the reconstructed ISAR image is obtained through the pre-trained HSNet network, so that the calculation complexity of the existing ISAR imaging problem is reduced; in addition, by setting the preset loss function as the negative variational lower bound, the imaging quality is improved, and the network interpretability is greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of radar data processing technology, specifically relating to a signal-to-noise ratio robust ISAR imaging method and device based on HSNet. Background Technology

[0002] Inverse Synthetic Aperture Radar (ISAR) plays a crucial role in space situational awareness and airborne target surveillance due to its all-weather, all-day, long-range, and high-resolution capabilities. However, in practical imaging missions, ISAR often faces complex observation conditions, such as changes in target distance leading to variations in signal-to-noise ratio (SNR). Therefore, achieving robust ISAR imaging under different SNR conditions is a primary research challenge for current technologies.

[0003] In existing technologies, sparse signal reconstruction methods transform the ISAR imaging problem into a sparse signal reconstruction problem through sparse models to optimize the solution. In recent years, deep network-based ISAR imaging methods have received close attention from the radar community. The main methods include data-driven methods and sparse Bayesian learning methods. Data-driven methods directly learn the relationship between input and labels through a driving network to obtain high-resolution images. However, due to the poor interpretability of the network, high-resolution ISAR imaging cannot be achieved in complex observation environments. Sparse Bayesian learning methods obtain the global optimum by learning statistical information about the target and environment. However, this method involves a large number of matrix inversion operations during maximum prediction and variational inference, leading to high computational complexity in the solution process.

[0004] Therefore, existing ISAR imaging methods suffer from poor imaging quality or high computational complexity. Summary of the Invention

[0005] To address the aforementioned problems in the prior art, this invention provides a signal-to-noise ratio robust ISAR imaging method and apparatus based on HSNet.

[0006] The technical problem to be solved by this invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention provides a signal-to-noise ratio robust ISAR imaging method based on HSNet, comprising:

[0008] Acquire the ISAR echo data to be imaged;

[0009] Motion compensation and range-Doppler imaging processing are performed on the ISAR echo data to be imaged to obtain the range-Doppler image of the original echo;

[0010] The original echo range-Doppler image is input into a pre-trained HSNet network, and the output is the parameter information of the reconstructed ISAR image;

[0011] The ISAR imaging results are obtained based on the parameter information of the reconstructed ISAR image; the preset loss function of the pre-trained HSNet network is a negative variational lower bound.

[0012] Optionally, the parameter information includes: the mean of the Gaussian distribution of the reconstructed ISAR image, the variance of the Gaussian distribution of the reconstructed ISAR image, the first parameter of the inverse gamma distribution of the noise variance, and the second parameter of the inverse gamma distribution of the noise variance.

[0013] Optionally, the step of performing motion compensation and range-Doppler imaging processing on the ISAR echo data to be imaged to obtain a range-Doppler image of the original echo includes:

[0014] The ISAR echo data to be imaged is sequentially subjected to envelope alignment and autofocus processing to complete motion compensation and obtain compensated ISAR echo data;

[0015] Range-Doppler imaging processing is performed on the compensated ISAR echo data to obtain the range-Doppler image of the original echo.

[0016] Optionally, obtaining the ISAR imaging result based on the parameter information of the reconstructed ISAR image includes:

[0017] The mean of the Gaussian distribution of the reconstructed ISAR image in the parameter information is used as the ISAR imaging result.

[0018] Optionally, the pre-trained HSNet network is a cardioid network structure; the cardioid network structure includes a first side processing structure and a second side processing structure.

[0019] The first side processing structure is used to fit the first parameter of the inverse gamma distribution of the noise variance and the second parameter of the inverse gamma distribution of the noise variance; the second side processing structure is used to fit the mean of the Gaussian distribution of the reconstructed ISAR image and the variance of the Gaussian distribution of the reconstructed ISAR image.

[0020] The first side processing structure and the second side processing structure are parameter-correlated through the preset loss function.

[0021] Optionally, the training process of the pre-trained HSNet network includes:

[0022] Acquire ISAR echo data samples;

[0023] Motion compensation and range-Doppler imaging processing are performed on the ISAR echo data samples to obtain range-Doppler image samples;

[0024] The initial HSNet network is trained using a preset loss function and the range Doppler image samples;

[0025] The initial HSNet network corresponding to the convergence condition of the preset loss function is used as the pre-trained HSNet network.

[0026] Optionally, the output of any node in the pre-trained HSNet network is represented as:

[0027]

[0028] Among them, o i,j Represents node X i,j The output of represents the number of downsampling operations, the number of convolutional layers, and the k-th convolutional layer. This represents multiple convolutional layers containing a leaky linear rectified activation function. This indicates a downsampling operation. This indicates an upsampling operation.

[0029] Optionally, the variational lower bound is expressed as:

[0030]

[0031] Indicates the variational lower bound. Represents distance Doppler image samples The likelihood function, This represents the difference between the variational posterior and the prior of the reconstructed ISAR image sample z. The variance of the sample noise σ 2 variational posterior and σ 2 The difference between prior knowledge;

[0032]

[0033]

[0034]

[0035] Represent z and σ 2 The presupposed conjugate prior, q(z, σ) 2 ) represents z and σ 2 The pre-defined global posterior distribution, It means that for q(z, σ) 2Find the expectation, where P represents the length of the distance matrix of the real image x, Q represents the length of the frequency domain matrix of the real image x, μ represents the mean of the Gaussian distribution of z, M represents the variance of the Gaussian distribution of z, and α represents σ. 2 The first parameter of the inverse gamma distribution, β represents σ. 2 The second parameter of the inverse gamma distribution, ψ represents the double gamma function, Tr represents the trace of the matrix, Φ represents the real-domain observation dictionary, and D KL Let represent the KL divergence processing, q(z) represent the pre-defined posterior distribution of z, p(z) represent the pre-defined conjugate Gaussian prior of z, U represent the width of the x-distance matrix, V represent the width of the x-frequency domain matrix, δ represent the hyperparameters of the dependency relationship between z and x, det(·) represents the calculation of the determinant of the input matrix, and q(σ 2 ) represents σ 2 The presupposed posterior distribution, p(σ) 2 ) represents σ 2 The predefined conjugate priors are: ρ represents the first hyperparameter, ω represents the second hyperparameter, Γ(·) represents the gamma function, and T represents the matrix transpose.

[0036] Secondly, the present invention provides a signal-to-noise ratio robust ISAR imaging device based on HSNet, the signal-to-noise ratio robust ISAR imaging device based on HSNet comprising: an acquisition unit, a compensation unit, and a calculation unit.

[0037] The acquisition unit is used to acquire the ISAR echo data to be imaged;

[0038] The compensation unit is used to perform motion compensation and range-Doppler imaging processing on the ISAR echo data to be imaged, so as to obtain the range-Doppler image of the original echo.

[0039] The computing unit is used to input the range-Doppler image of the original echo into a pre-trained HSNet network and output parameter information for the reconstructed ISAR image.

[0040] The acquisition unit is further configured to acquire ISAR imaging results based on the parameter information of the reconstructed ISAR image; the preset loss function of the pre-trained HSNet network is a negative variational lower bound.

[0041] Thirdly, the present invention provides a signal-to-noise ratio robust ISAR imaging device based on HSNet, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the HSNet-based signal-to-noise ratio robust ISAR imaging device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the HSNet-based signal-to-noise ratio robust ISAR imaging method as described in the first aspect above.

[0042] This invention provides a signal-to-noise ratio (SNR) robust ISAR imaging method and apparatus based on HSNet. The SNR robust ISAR imaging method based on HSNet includes: acquiring ISAR echo data to be imaged; performing motion compensation and range-Doppler imaging processing on the ISAR echo data to obtain a range-Doppler image of the original echo; inputting the range-Doppler image of the original echo into a pre-trained HSNet network, outputting parameter information for reconstructing the ISAR image; obtaining the ISAR imaging result based on the parameter information of the reconstructed ISAR image; and setting the preset loss function of the pre-trained HSNet network to a negative variational lower bound. In this invention, by acquiring the parameter information of the reconstructed ISAR image through a pre-trained HSNet network, the frequent matrix inversions in traditional Bayesian learning are effectively avoided, reducing the computational complexity of existing ISAR imaging problems. Furthermore, by setting the preset loss function to a negative variational lower bound, not only is the imaging quality improved, but the network's interpretability is also greatly enhanced because this preset loss function can accurately estimate the noise variance.

[0043] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0044] Figure 1 A schematic flowchart of a signal-to-noise ratio robust ISAR imaging method based on HSNet provided in an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the structure of a pre-trained HSNet network provided in an embodiment of the present invention;

[0046] Figure 3 The above describes the contrast imaging results at a signal-to-noise ratio of 5dB provided in this embodiment of the invention.

[0047] Figure 4 The above describes the contrast imaging results at a signal-to-noise ratio of 15 dB provided in this embodiment of the invention.

[0048] Figure 5 A schematic diagram of a signal-to-noise ratio robust ISAR imaging device based on HSNet is provided for an embodiment of the present invention;

[0049] Figure 6 This is a schematic diagram of a signal-to-noise ratio robust ISAR imaging device based on HSNet, provided as an embodiment of the present invention. Detailed Implementation

[0050] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0051] To reduce the computational complexity of ISAR imaging while improving imaging quality, this invention provides a signal-to-noise ratio robust ISAR imaging method based on HSNet. Figure 1 A flowchart illustrating a signal-to-noise ratio robust ISAR imaging method based on HSNet, as provided in an embodiment of the present invention, is shown below. Figure 1 As shown, it includes:

[0052] S101. Acquire the ISAR echo data to be imaged.

[0053] S102. Perform motion compensation and range-Doppler imaging processing on the ISAR echo data to be imaged to obtain the range-Doppler image of the original echo.

[0054] Optionally, S102 may specifically include:

[0055] The ISAR echo data to be imaged is sequentially envelope aligned and autofocused to complete motion compensation and obtain compensated ISAR echo data.

[0056] Range-Doppler imaging processing was performed on the compensated ISAR echo data to obtain the range-Doppler image of the original echo.

[0057] S103. Input the original echo range-Doppler image into the pre-trained HSNet network and output the parameter information of the reconstructed ISAR image.

[0058] Optionally, the parameter information includes: the mean of the Gaussian distribution of the reconstructed ISAR image, the variance of the Gaussian distribution of the reconstructed ISAR image, the first parameter of the inverse gamma distribution of the noise variance, and the second parameter of the inverse gamma distribution of the noise variance.

[0059] Optionally, the pre-trained HSNet network is a cardioid network structure; the cardioid network structure includes a first-side processing structure and a second-side processing structure.

[0060] The first side processing structure is used to fit the first parameter of the inverse gamma distribution of the noise variance and the second parameter of the inverse gamma distribution of the noise variance; the second side processing structure is used to fit the mean of the Gaussian distribution of the reconstructed ISAR image and the variance of the Gaussian distribution of the reconstructed ISAR image.

[0061] The first-side processing structure and the second-side processing structure are correlated by a preset loss function.

[0062] Optionally, the training process of the pre-trained HSNet network includes:

[0063] Acquire ISAR echo data samples;

[0064] Motion compensation and range-Doppler imaging processing were performed on the ISAR echo data samples to obtain range-Doppler image samples.

[0065] The initial HSNet network is trained using a preset loss function and range Doppler image samples;

[0066] The initial HSNet network corresponding to the convergence condition of the preset loss function is used as the pre-trained HSNet network.

[0067] Optionally, the output of any node in the pre-trained HSNet network is represented as:

[0068]

[0069] Among them, o i,j Represents node X i,j The output of represents the number of downsampling operations, the number of convolutional layers, and the k-th convolutional layer. This represents multiple convolutional layers containing a leaky linear rectified activation function. This indicates a downsampling operation. This indicates an upsampling operation.

[0070] Figure 2 This is a schematic diagram of the structure of a pre-trained HSNet network provided in an embodiment of the present invention. Figure 2 As shown, the outputs μ and M on the right can be used to finally fit the mean and variance of the Gaussian distribution of the reconstructed ISAR image, respectively. α and β on the left can be used to finally fit the first and second parameters of the inverse gamma distribution of the noise variance, respectively. A loss function can also be used... The data from the left and right sides are correlated, and the loss function is a negative number of the variational lower bound.

[0071] Additionally, at each node X i,j Convolution processing is performed at the corresponding location, X i,j The kernel sizes at each location are shown in Table 1.

[0072] Table 1. Node X i,j kernel size at the point

[0073] node <![CDATA[X 0,-4～4 ]]> <![CDATA[X 1,-3～3 ]]> <![CDATA[X 2,-2～2 ]]> <![CDATA[X 3,-1～1 ]]> <![CDATA[X 4,0 ]]> 32 64 128 256 512

[0074] Optionally, the variational lower bound is expressed as:

[0075]

[0076] Indicates the variational lower bound. Represents distance Doppler image samples The likelihood function, This represents the difference between the variational posterior and the prior of the reconstructed ISAR image sample z. The variance of the sample noise σ 2 variational posterior and σ 2 The difference between prior knowledge;

[0077]

[0078]

[0079]

[0080] Represent z and σ 2 The presupposed conjugate prior, q(z, σ) 2 ) represents z and σ 2 The pre-defined global posterior distribution, It means that for q(z, σ) 2 Find the expectation, where P represents the length of the distance matrix of the real image x, Q represents the length of the frequency domain matrix of the real image x, μ represents the mean of the Gaussian distribution of z, M represents the variance of the Gaussian distribution of z, and α represents σ. 2 The first parameter of the inverse gamma distribution, β represents σ. 2 The second parameter of the inverse gamma distribution, ψ represents the double gamma function, Tr represents the trace of the matrix, Φ represents the real-domain observation dictionary, and D KL Let represent the KL divergence processing, q(z) represent the pre-defined posterior distribution of z, p(z) represent the pre-defined conjugate Gaussian prior of z, U represent the width of the x-distance matrix, V represent the width of the x-frequency domain matrix, δ represent the hyperparameters of the dependency relationship between z and x, det(·) represents the calculation of the determinant of the input matrix, and q(σ 2 ) represents σ 2 The presupposed posterior distribution, p(σ) 2 ) represents σ 2 The predefined conjugate priors are: ρ represents the first hyperparameter, ω represents the second hyperparameter, Γ(·) represents the gamma function, and T represents the matrix transpose.

[0081] From the above formula, it can be seen that, Influenced by parameters μ and M, Influenced by parameters α and β, and It is simultaneously influenced by μ, M, α, and β. Furthermore, when the variational lower bound reaches its maximum value, the posterior distribution q(z,σ) 2 It infinitely approximates the true posterior distribution. At this point, the parameters of the HSnet network used in this invention are obtained to the optimal solution. Therefore, this invention proposes to use the negative variational lower bound as the loss function of the HSnet network.

[0082] In this embodiment of the invention, the construction process of the pre-trained HSNet network and the formulation of the preset loss function are derived and explained:

[0083] In this embodiment of the invention, the complex domain distance-Doppler image of the original echo... and the true distribution of scattering points in the complex domain The relationship between them is:

[0084]

[0085] in, It is a radar echo data range dictionary. This is the radar echo data azimuth dictionary, where n0' represents complex domain noise. Representing the complex field, P and Q represent the length of the matrix, and U and V represent the width of the matrix. This represents the two-dimensional inverse Fourier transform. Let Kronecker product be represented. For ease of Bayesian inference, formula (1) is expressed in the real number field as follows:

[0086]

[0087] Where Re(·) represents the real part and Im(·) represents the imaginary part, the sparse observation model satisfies:

[0088]

[0089] in Φ represents the real-domain distance-Doppler image (distance of the original echo - Doppler image), x' represents the real-domain true scattering point center distribution, n' represents the real-domain noise, and Φ represents the real-domain observation dictionary.

[0090] Because Bayesian inference models are more accurate in statistically analyzing noise distributions, to better remove noise, the sparse observation model is constructed as a Bayesian inference model. Given prior information about the imaging results and prior information about the noise, the posterior distributions of the imaging results and noise are derived. Specifically:

[0091] As can be seen from the sparse observation model, when the noise is Gaussian noise with a mean of 0, It also follows a Gaussian distribution, expressed as:

[0092]

[0093] Indicate z' and σ 2' The pre-defined conjugate prior, z' represents the reconstructed ISAR image, σ 2'Let I represent the noise variance, and let I represent the identity matrix.

[0094] In fact, the distribution of the true scattering point center x' has strong prior information about the reconstructed distribution z'. Considering this factor, z' can be set as a conjugate Gaussian prior:

[0095]

[0096] δ' represents the hyperparameter of the dependency relationship between z' and x'.

[0097] Assume that the conjugate prior of the noise variance follows an inverse gamma distribution:

[0098] p(σ 2' )=IG(σ 2' |a,b); (6)

[0099] a represents the first hyperparameter of the inverse gamma distribution, and b represents the second hyperparameter of the inverse gamma distribution.

[0100] Combining the above formulas 4-6, we can obtain that the posterior distribution of z' satisfies:

[0101]

[0102] However, due to the posterior distribution of z' It is difficult to solve, so a posterior distribution q(z', σ) is artificially constructed. 2' ),right To approximate the result, based on the "mean field" assumption, q(z', σ) can be approximated. 2' Decompose the following to obtain:

[0103] q(z',σ 2' )=q(z')q(σ 2' (8)

[0104] set up:

[0105]

[0106] q(σ 2' )=IG(σ 2' |α',β');(10)

[0107] Based on this, the problem is transformed into solving for μ', M', α', and β'. μ' represents the mean of the Gaussian distribution of the reconstructed ISAR image, M' represents the variance of the Gaussian distribution of the reconstructed ISAR image, α' represents the first parameter of the inverse gamma distribution of the noise variance, and β' represents the second parameter of the inverse gamma distribution of the noise variance. Based on this, this embodiment of the invention introduces the HSNet network for parameter fitting of μ', M', α', and β'.

[0108] It should be noted that by performing sparse Bayesian modeling under different signal-to-noise ratios and echo loss conditions, the variance of the reconstructed image and noise is modeled as Gaussian and inverse gamma distributions, and their posterior distributions are derived. Then, a pre-trained HSNet network is used to fit the posterior distribution parameters, ultimately obtaining the ISAR imaging results, which improves the imaging quality and the interpretability of the network.

[0109] S104. Obtain ISAR imaging results based on parameter information from the reconstructed ISAR image.

[0110] The default loss function of the pre-trained HSNet network is a negative variational lower bound.

[0111] This invention provides a signal-to-noise ratio robust ISAR imaging method based on HSNet, comprising: acquiring ISAR echo data to be imaged; performing motion compensation and range-Doppler imaging processing on the ISAR echo data to obtain a range-Doppler image of the original echo; inputting the range-Doppler image of the original echo into a pre-trained HSNet network, and outputting parameter information for reconstructing the ISAR image; obtaining the ISAR imaging result based on the parameter information of the reconstructed ISAR image; the preset loss function of the pre-trained HSNet network is a negative variational lower bound. In this invention, by acquiring the parameter information of the reconstructed ISAR image through a pre-trained HSNet network, the frequent matrix inversion in traditional Bayesian learning is effectively avoided, reducing the computational complexity of existing ISAR imaging problems; secondly, by setting the preset loss function to a negative variational lower bound, not only is the imaging quality improved, but also the interpretability of the network is greatly improved because the preset loss function can accurately estimate the noise variance.

[0112] Optionally, S104 may specifically include:

[0113] The mean of the Gaussian distribution of the reconstructed ISAR image in the parameter information is used as the ISAR imaging result.

[0114] To illustrate the effectiveness of the HSNet-based signal-to-noise ratio robust ISAR imaging method provided in this embodiment of the invention, simulation experiments were also conducted. These experiments compared the final imaging results of three methods—2D-AND, UNet, and the method of this invention—on the same measured data. Figure 3 This is a contrast imaging result with a signal-to-noise ratio of 5dB provided in an embodiment of the present invention. Figure 4 This is a contrast imaging result at a signal-to-noise ratio of 15dB provided in an embodiment of the present invention. Figure 3 and Figure 4 Figure (a) shows the imaging results of the 2D-AND method. Figure 3 and Figure 4 Figure (b) shows the imaging results of the UNet method. Figure 3and Figure 4 Figure (c) shows the imaging results of the method of the present invention. From... Figure 3 and Figure 4 It can be seen that the method of the present invention has the best imaging results under different signal-to-noise ratios, which verifies the effectiveness of the method of the present invention.

[0115] The method provided in this embodiment of the invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc., and this embodiment of the invention does not limit the application to such devices.

[0116] Based on the same inventive concept, embodiments of the present invention also provide a signal-to-noise ratio robust ISAR imaging device based on HSNet. Figure 5 This is a schematic diagram of a signal-to-noise ratio robust ISAR imaging device based on HSNet, provided as an embodiment of the present invention. Figure 5 As shown, a signal-to-noise ratio robust ISAR imaging device based on HSNet includes: an acquisition unit 501, a compensation unit 502, and a calculation unit 503.

[0117] The acquisition unit 501 is used to acquire the ISAR echo data to be imaged;

[0118] The compensation unit 502 is used to perform motion compensation and range-Doppler imaging processing on the ISAR echo data to be imaged, so as to obtain the range-Doppler image of the original echo.

[0119] The computing unit 503 is used to input the distance-Doppler image of the original echo into the pre-trained HSNet network and output the parameter information of the reconstructed ISAR image.

[0120] The acquisition unit 501 is also used to acquire ISAR imaging results based on the parameter information of the reconstructed ISAR image; the preset loss function of the pre-trained HSNet network is a negative variational lower bound.

[0121] Figure 6 A schematic diagram of a signal-to-noise ratio robust ISAR imaging device based on HSNet, provided in an embodiment of the present invention, includes: a processor 710, a storage medium 720, and a bus 730. The storage medium 720 stores machine-readable instructions executable by the processor 710. When the HSNet-based signal-to-noise ratio robust ISAR imaging device is running, the processor 710 and the storage medium 720 communicate via the bus 730. The processor 710 executes the machine-readable instructions to perform the steps of the above-described method embodiment. Specific implementation methods and technical effects are similar and will not be repeated here.

[0122] The storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the storage medium may also be at least one storage device located remotely from the aforementioned processor.

[0123] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0124] The present invention also provides a computer-readable storage medium. A computer program is stored in the computer-readable storage medium, and when executed by a processor, the computer program implements the steps of any of the above-described HSNet-based signal-to-noise ratio robust ISAR imaging methods.

[0125] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.

[0126] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0127] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0128] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A signal-to-noise ratio robust ISAR imaging method based on HSNet, characterized in that, include: Acquire the ISAR echo data to be imaged; Motion compensation and range-Doppler imaging processing are performed on the ISAR echo data to be imaged to obtain the range-Doppler image of the original echo; The original echo range-Doppler image is input into a pre-trained HSNet network, and the output is the parameter information of the reconstructed ISAR image; The ISAR imaging result is obtained based on the parameter information of the reconstructed ISAR image; the preset loss function of the pre-trained HSNet network is a negative variational lower bound; The variational lower bound is expressed as: ； Indicates the variational lower bound. Represents distance Doppler image samples The likelihood function, Representative ISAR image sample variational posterior and The difference between the priors, Represents the sample noise variance variational posterior and The difference between prior knowledge; ； ； ； express and The presupposed conjugate prior, express and The pre-defined global posterior distribution, Indicates to Seeking expectations, Represents real images The length of the distance dimension matrix, Represents real images The length of the frequency domain matrix, express The mean of the Gaussian distribution, express The variance of the Gaussian distribution, express The first parameter of the inverse gamma distribution, express The second parameter of the inverse gamma distribution, Represents the double gamma function. Represents the trace of a matrix. Represents a dictionary of observations in the real number field. This indicates the KL divergence treatment. express The pre-defined posterior distribution, express The presupposed conjugate Gaussian prior, express The width of the distance dimension matrix, represent The width of the frequency domain matrix, express and Hyperparameters of dependency relationships This indicates the calculation of the determinant of the input matrix. express The pre-defined posterior distribution, express The presupposed conjugate prior, Indicates the first hyperparameter. Indicates the second hyperparameter. Let T denote the gamma function, and T denote the matrix transpose.

2. The signal-to-noise ratio robust ISAR imaging method based on HSNet according to claim 1, characterized in that, The parameter information includes: the mean of the Gaussian distribution of the reconstructed ISAR image, the variance of the Gaussian distribution of the reconstructed ISAR image, the first parameter of the inverse gamma distribution of the noise variance, and the second parameter of the inverse gamma distribution of the noise variance.

3. The signal-to-noise ratio robust ISAR imaging method based on HSNet according to claim 1, characterized in that, The process of performing motion compensation and range-Doppler imaging processing on the ISAR echo data to be imaged to obtain the range-Doppler image of the original echo includes: The ISAR echo data to be imaged is sequentially subjected to envelope alignment and autofocus processing to complete motion compensation and obtain compensated ISAR echo data; Range-Doppler imaging processing is performed on the compensated ISAR echo data to obtain the range-Doppler image of the original echo.

4. The signal-to-noise ratio robust ISAR imaging method based on HSNet according to claim 2, characterized in that, The process of obtaining ISAR imaging results based on the parameter information of the reconstructed ISAR image includes: The mean of the Gaussian distribution of the reconstructed ISAR image in the parameter information is used as the ISAR imaging result.

5. The signal-to-noise ratio robust ISAR imaging method based on HSNet according to claim 2, characterized in that, The pre-trained HSNet network has a cardioid network structure; the cardioid network structure includes a first side processing structure and a second side processing structure. The first side processing structure is used to fit the first parameter of the inverse gamma distribution of the noise variance and the second parameter of the inverse gamma distribution of the noise variance; the second side processing structure is used to fit the mean of the Gaussian distribution of the reconstructed ISAR image and the variance of the Gaussian distribution of the reconstructed ISAR image. The first side processing structure and the second side processing structure are parameter-correlated through the preset loss function.

6. The signal-to-noise ratio robust ISAR imaging method based on HSNet according to claim 1, characterized in that, The training process of the pre-trained HSNet network includes: Acquire ISAR echo data samples; Motion compensation and range-Doppler imaging processing are performed on the ISAR echo data samples to obtain range-Doppler image samples; The initial HSNet network is trained using a preset loss function and the range Doppler image samples; The initial HSNet network corresponding to the convergence condition of the preset loss function is used as the pre-trained HSNet network.

7. The signal-to-noise ratio robust ISAR imaging method based on HSNet according to claim 1, characterized in that, The output of any node in the pre-trained HSNet network is represented as follows: ； in, Represents a node The output, Indicates the number of downsampled samples. Indicates the number of convolutional layers. Indicates the first One convolutional layer, This represents multiple convolutional layers containing a leaky linear rectified activation function. This indicates a downsampling operation. This indicates an upsampling operation.

8. A signal-to-noise ratio robust ISAR imaging device based on HSNet, characterized in that, The HSNet-based signal-to-noise ratio robust ISAR imaging device includes: an acquisition unit, a compensation unit, and a calculation unit; The acquisition unit is used to: acquire ISAR echo data to be imaged; The compensation unit is used to: perform motion compensation and range-Doppler imaging processing on the ISAR echo data to be imaged, and obtain the range-Doppler image of the original echo; The computing unit is used to: input the distance-Doppler image of the original echo into a pre-trained HSNet network, and output the parameter information of the reconstructed ISAR image; The acquisition unit is further configured to: acquire ISAR imaging results based on the parameter information of the reconstructed ISAR image; the preset loss function of the pre-trained HSNet network is a negative variational lower bound, and the variational lower bound is expressed as: ； Indicates the variational lower bound. Represents distance Doppler image samples The likelihood function, Representative ISAR image sample variational posterior and The difference between the priors, Represents the sample noise variance variational posterior and The difference between prior knowledge; ； ； ； express and The presupposed conjugate prior, express and The pre-defined global posterior distribution, Indicates to Seeking expectations, Represents real images The length of the distance dimension matrix, Represents real images The length of the frequency domain matrix, express The mean of the Gaussian distribution, express The variance of the Gaussian distribution, express The first parameter of the inverse gamma distribution, express The second parameter of the inverse gamma distribution, Represents the double gamma function. Represents the trace of a matrix. Represents a dictionary of observations in the real number field. This indicates the KL divergence treatment. express The pre-defined posterior distribution, express The presupposed conjugate Gaussian prior, express The width of the distance dimension matrix, represent The width of the frequency domain matrix, express and Hyperparameters of dependency relationships This indicates the calculation of the determinant of the input matrix. express The pre-defined posterior distribution, express The presupposed conjugate prior, Indicates the first hyperparameter. Indicates the second hyperparameter. Let T denote the gamma function, and T denote the matrix transpose.

9. A signal-to-noise ratio robust ISAR imaging device based on HSNet, characterized in that, include: The device includes a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the HSNet-based signal-to-noise ratio robust ISAR imaging device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the HSNet-based signal-to-noise ratio robust ISAR imaging method as described in any one of claims 1-7.

Images